I've been building agents for a couple years and context engineering keeps being the bottleneck. Curious what problems others are hitting.

Things I've struggled with:

- Debugging what the agent actually saw at decision time

- Managing context across multi-agent patterns

- Keeping history without blowing up storage

What's eating your time?